Saving history with AI. Advances in nerve regeneration. Tricking AI into believing a fake disease. All that and more in this week’s edition of The Prototype. All that and more in this week’s edition of The Prototype. To get it in your inbox, sign up here.

All around the country, historical societies, libraries and universities have boxes of archived material waiting to be reviewed and cataloged. It can take an archivist more than an hour to process the items in a single box. As a result, some institutions have months-long or even years-long backlogs, keeping key parts of our history unknown to the present.

Dean Serrentino and his company, Historiq, aim to fix this bottleneck with Una, an AI platform designed to accelerate archiving. Instead of painstakingly cataloging items “with a clipboard and a pencil,” Una enables archivists to verbally describe their observations as they examine material, Serrentino said. Those notes are then plugged into their institution’s data system. Plus, any documents they find can be simultaneously digitized using laptop or phone cameras.

This both speeds up and improves accuracy of the cataloging process. When they don’t have to write everything down, archivists put “more context and historical description out of their brains and into the system,” he said. He added that the AI only produces drafts that require review and approval to ensure the human element is present.

Historiq was founded in 2025 with a $1.25 million investment from edtech company Curriculum Associates Chairman Rob Waldron. It already counts several institutions among its customers. One of those is Revolutionary War site Fort Ticonderoga, which plans to use Una to digitize its rare book collection later this year.

Discovery of the Week: Tricking AI Into Believing In A Fake Disease

A team of scientists invented a condition called “bixonimania” and published two fake papers about it on a preprint server in early 2024, along with a couple of posts on Medium. The papers were published under the name Lazljiv Izgubljenovic, a made-up researcher, who was said to work at Asteria Horizon University, a non-existent university in the fictional Nova City, California.

Within weeks, according to Nature, major AI chatbots started reporting on the “disease” as though it were real–and the fake studies started getting cited in peer-reviewed papers, suggesting that scientists were relying on AI for background research while preparing their own studies.

Large language models incorporated these papers into the data despite the fact that they deliberately contained multiple flags, such as the phrase “this entire paper is made up” and claimed the “research” was funded by the “Professor Sideshow Bob Foundation for its work in advanced trickery.”

Google Gemini, Microsoft Copilot, Perplexity AI and OpenAI’s ChatGPT all readily provided information about the fake disease, according to the researchers. That said, more recent model versions had more nuanced takes of the condition.

The purpose of the experiment was to see how readily AI chatbots would use misinformation as the basis of legitimate health advice to users. This builds on previous research from the team's leader, Almira Osmanovic Thunström, which found that LLMs were more likely to accept false information that was presented as an academic paper.

This experiment reminds of a more informal one conducted by BBC tech journalist Thomas Germain, who was able to get major AI chatbots to report that he was a hot dog eating champion with just a single well-crafted blog post. But both of these examples highlight that even artificial intelligence isn’t immune to one of the iron laws of computing: Garbage In, Garbage Out.

NervGen Advances Its Spinal Cord Injury Treatments

Last December, I talked to Adam Rogers, whose company NervGen is developing a drug that helps injured spinal cords heal. In a small clinical trial, the treatment improved hand function, lower body mobility and upper limb strength in patients with spinal injuries. Now the company is moving on to the next step. It just reached alignment with the FDA on a design for a larger clinical trial, this time with 150 patients at up to 60 different sites in the U.S. and Canada.

The new study, Rogers told me, will start this summer and hopefully end in the second half of 2027. If this larger trial shows similar results to their first, the company hopes to begin the FDA approval process for their drug in 2028. In the meantime, the company plans to release more data this year in a study of how well their drug is improving spinal cord patient outcomes, and is also looking to announce a clinical study for another disease the company’s neuroreparative treatments might help with.

The Hot Take: We Should Be Talking About Generative Science

Each week, I ask investors for their take on tech trends within their industries. Today I’m featuring thoughts from James Joaquin, cofounder and managing partner at Obvious Ventures. He looks to make early investments in breakthrough technologies such as advanced materials, artificial intelligence and plant-based protein.

What tech is being overhyped right now?

Humanoid Robots. I believe the technology around bipedal humanoid robots is being overhyped. The real unsolved problem in robotics is hand dexterity and fine motor skills. Solving that will unlock a multi-trillion dollar market for robotics without requiring legs or mobility for robots. Building superhuman hands will be the ChatGPT moment for robotics.

What tech should more people be talking about today?

Generative Science. Let’s talk more about the positive impact AI is having on human health. Since the Nobel-prize winning work of AlphaFold, we are seeing new breakthroughs in health and pharma driven by modern AI. Hospitals are using AI to identify risk and early detection of infections and saving lives in the process. Researchers are using AI co-scientists and open-source biomolecular interaction models like Boltz to dramatically accelerate the development of new cures.

What tech are we all going to be talking about in five years?

Personal AI. In five years, each of us will have a private, trusted AI assistant. We’ll be talking to our human doctors about health insights and diagnostics surfaced by our AI, talking to our human coworkers about business strategy co-authored by our AI, and hopefully spending more time with the people we love thanks to the tasks outsourced to our AI.

Anthropic vs The Pentagon, Ctd: Last month, a federal judge placed a stay on the Pentagon’s designation of Anthropic as a supply chain risk. But this week, an appeals court overturned that ruling. It means that at least while the case is being considered by the courts, defense contractors won’t be able to use Anthropic’s software on any work they do for the federal government.

Space imaging moves: A number of announcements this week highlight that the business of Earth imaging is still alive and well. Geospatial data company Xoople announced a partnership with L3Harris to develop a satellite constellation whose data outputs are optimized for AI models. Meanwhile, imaging company Vantor announced it’s building a satellite constellation that’s capable of both high-resolution imaging and continuous monitoring–a challenging prospect because usually focusing on one means trading off capability for the other. Finally, Firefly Aerospace announced a deal with Nvidia to provide the processing power for Ocula, its project to develop a commercial Moon imaging and mapping system.

What’s Entertaining Me This Week

Reviving a sitcom is always a dicey prospect–it’s hard to recapture the old magic and chemistry that made the original. So it’s pretty impressive that Scrubs was able to pull it off. The original core cast has slid right back into their old roles, believably playing older, somewhat-wiser versions of themselves, and the new additions are fantastic. Especially Joel Kim Booster, who steals every single scene he’s in as Dr. Kevin Park. If you were a fan of the show, you won’t be disappointed.

© Forbes